.png){kind=icon}
.png){kind=icon}
Return
Model Lightweight Strategy Under Federated Edge Learning Framework
.png){kind=icon}
PRE
AI
DOI：10.1109/TVT.2025.3604824.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Mobile communication devices such as mobile phones and autonomous driving in car systems are ubiquitous.The energy and computing power of existing mobile devices are limited. The energy consumption of local training in federated learning is significant. To reduce computational energy consumption, we propose a lightweight federated edge learning framework for client models. We attempted to extract a small model suitable for execution on mobile devices from a large federated global model trained on edge servers. Directly training the small model would involve multiple transmissions of local models, leading to a shift of energy consumption from computation to communication. We aim to transfer more computation to edge servers. To this end, we design a pseudo vector compression method to reduce data transmission energy consumption. The framework enables most model training to be moved to edge servers while also ensuring the accuracy and reducing the energy consumption of client models. In addition, we prune and adjust the model parameters of the lightweight model to make it more personalized. We also provided a convergence analysis of our algorithm and proved that our algorithm is convergent. The experimental results indicate that our proposed framework and algorithm not only reduce computational costs of mobile devices but also ensure the availability of local lightweight models.
Keywords:
Federated edge learning
lightweight model
pruning
pseudo vector sharing
computational energy optimization
Journal
IF:
7.1
Papers:
1.8W
Citations:
6.6W